feat(hub): services-index landing page at /

[growlf]

Closes the operator-friction Garth flagged: "a single entry point that links all of our resource pages on these deployments into a single web UI for easier access by humans. Memorizing all of the ports and manually typing them all in is tedious."

What changed

New: shepherd/ui/hub.html

Landing page rendered at GET / on shepherd-control (port 40117). Three sections:

1. Quick-link cards for every service on this host:

   • Shepherd herd (this app's /herd dashboard) → :40117/herd
   • Olla federation + load-balancer → :40114/
   • Smart Router gestalt UI → :40115/gestalt/ui
   • LiteLLM admin UI → :4000/ui
   • Retriever health → :42000/health
   • Ollama API → :11434/api/tags

   Each card shows: emoji + name + short description, live up/down
   status badge (refreshed every 15s), port number, and an Open ↗ link.

   Auto-discovers the host hostname from window.location, so links
   work from localhost, LAN IP, mesh hostname, or VPN-routed access —
   no template substitution required.

2. Live status banner: 'N / M services up' (refreshes with the cards).

3. Documentation + hardware guides section linking to:

   • README, Troubleshooting, Retriever guide
   • Intel iGPU Vulkan guide + per-hardware docs index
   • growlf/intel_nuc_skullcanyon_ollama_with_gpu standalone repo (per Garth's per-hardware-target OSS-repo doctrine)
   • ai-stack GitHub repo + open issues + CONTRIBUTING

New: shepherd/ui/hub.js

• Renders the service cards from a small SERVICES catalog
• Polls each service's lightweight health endpoint every 15s using fetch(... mode: 'no-cors'); opaque-response = up, throw/timeout = down
• 3 sec per-probe timeout, runs in parallel via Promise.all
• Updates the overall-status banner with up/total count

Modified: shepherd/ui/shepherd.css

Appended hub-page styles (~165 lines) under a 'body.hub' selector, leaving the existing kiosk (body.kiosk) styling untouched:

• .hub-grid: auto-fill card-grid with min 280px columns
• .service-card: dark theme matching kiosk; colored left-border by status
• .service-badge: pill-shaped up/down/checking indicator
• .doc-card: documentation section card grid

Modified: shepherd/shepherd_control/app.py

Two route changes (no other logic touched):

• GET / → now serves hub.html (was: index.html)
• GET /herd → new; serves index.html (the existing kiosk herd view)

Static-file mount at /ui/* is unchanged. /herd/aggregate JSON data endpoint is unchanged. The kiosk view JS still works at /herd — it fetches /herd/aggregate as before.

Test plan

• Python syntax: python3 -m py_compile shepherd/shepherd_control/app.py
• JS parse: node --check shepherd/ui/hub.js
• HTML tag balance verified
• Live: GET / returns hub.html with service cards
• Live: GET /herd returns existing kiosk view, no regression
• Live: status badges update correctly for running vs stopped services
• Live: 'Open ↗' links use the same hostname the user accessed by

Scope notes

Per scope-decision (Medium) on PR #41 / d758e57d:

• IN scope: hub-page + status badges + docs section
• OUT of scope (separate PRs): LiteLLM admin login fix; on-cluster substrate-verify Shepherd shows all peers / dashboard-redraw issue. Those are observability-backlog items that fold into followup work.

Summary

Type of change

• Bug fix
• New feature / enhancement
• Documentation update
• Refactor / cleanup
• CI / tooling change

Related issues

Testing

• docker compose config passes with no errors
• Shell scripts pass shellcheck
• Manually tested on Intel Arc hardware (if relevant)

Checklist

• My changes follow the project's code style
• I have updated documentation where necessary
• I have not committed secrets or real credentials
• .env.example is up to date (if env vars were added/removed)